Predicting high-frequency returns using order book data
Konstantinos Vafeidis, Associate (London)
![Post feature](/rails/active_storage/representations/proxy/eyJfcmFpbHMiOnsibWVzc2FnZSI6IkJBaHBBdFJPIiwiZXhwIjpudWxsLCJwdXIiOiJibG9iX2lkIn19--1fe8121a970c995fc6e2034b373bbe25e60f7e70/eyJfcmFpbHMiOnsibWVzc2FnZSI6IkJBaDdCem9MWm05eWJXRjBTU0lJY0c1bkJqb0dSVlE2QzNKbGMybDZaVWtpRFRFeU9EQjRPRFV6QmpzR1ZBPT0iLCJleHAiOm51bGwsInB1ciI6InZhcmlhdGlvbiJ9fQ==--4a0ea5eeb6f0e6d47bff3b653fd69e214040023f/zac-ong-ZCGn3SVR8UA-unsplash.jpg)
The race to zero latency has seen vast technology investments from high-frequency traders, but few firms are willing to reveal their secrets. In this literature review, we summarise a paper that uses machine learning and order book data to show how ultra-low latency strategies produce returns.
QUICK VIEW
- The authors use the Nasdaq TotalView ITCH order book to conduct their experiments.
- Predictability in price formation dynamics is strong up to 50-300 order book updates ahead, or up to nearly half a second ahead.
- Predictability is easier to identify in high-frequency trading strategies than in low-frequency strategies, although it is easier to trade once discovered in the latter. This is because of technological limitations in high-frequency market structure.